Skip to content
Odysı
See if we fit
← Guides
Guide Data and privacy

What happens to your data when you use AI

It is one of the questions we hear most from businesses: what happens to our data, and our customers’ data, when an AI reads it? Is it used for training? Does it leave the EU? Should the model run on our own server? And what about the client file someone pasted into a personal chat app? Here are the answers as the providers’ own terms stood on 19 November 2026.

T
Updated 19 November 2026
Topic Data and privacy
Read 8 min
No. 01

Three ways your data reaches an AI

  • A person pastes it into a chat app on their own account, under consumer terms the company never agreed to.
  • A business plan the company manages: ChatGPT Business or Enterprise, Claude Team or Enterprise, Gemini in Google Workspace.
  • An AI agent built on the API, the route we use. It reads only what it is connected to, under the provider’s commercial terms.

The plan matters more than the brand, and paying does not make a plan a business plan. Anthropic lists Free, Pro, Max and Ultra as consumer plans; Team, Enterprise and the API are commercial. OpenAI separates “Services for individuals, such as ChatGPT” from its products for business users.

No. 02

Training and retention, provider by provider

What each provider’s pages said on 19 November 2026:

  • Anthropic (Claude). On consumer plans, chats are used to improve its models if you choose to allow it, and then kept in de-identified form for up to 5 years in its training pipelines. Commercial products: “By default, we will not use your inputs or outputs from our commercial products […] to train our models” (privacy centre), unless you send feedback or opt in. API inputs and outputs are deleted within 30 days, unless you agree zero data retention or content is flagged under its Usage Policy.
  • OpenAI (ChatGPT). On services for individuals, OpenAI may use your content to train its models unless you opt out; Temporary Chat is not used for training. “By default, we do not train on any inputs or outputs from our products for business users, including ChatGPT Team, ChatGPT Enterprise, and the API.” API logs are kept up to 30 days for abuse monitoring; zero data retention needs OpenAI’s prior approval.
  • Google (Gemini). In the Gemini app on a personal account, with the history setting on, Google trains on your future chats, and chats seen by reviewers are kept up to three years. In Google Workspace, your content is not reviewed by people or used for training outside your domain without permission (Workspace privacy hub). On the Gemini API, unpaid use improves Google’s products and may be read by human reviewers; paid use does not, and apps for users in the EEA, Switzerland or the UK must use paid services.

For anyone negotiating zero data retention: since 9 June 2026, Anthropic keeps prompts and outputs for 30 days for its most capable models (the Mythos class and future models it designates), on every platform; organisations with zero data retention must accept that retention to use them. OpenAI reserves a similar right for specific models.

No. 03

Where the data is processed

On Anthropic’s own API you can choose where the model runs, but only between the US and global routing, and data at rest is stored in the US (data residency docs). Processing inside the EU is offered through the regional endpoints of Amazon Bedrock and Google Cloud’s Vertex AI, which keep storage and processing in the region (not through Anthropic’s own offerings on AWS or Google Cloud, which store data in the US). So nobody should promise you that Claude, called directly, never sends data to the US.

OpenAI offers storage in Europe (EEA and Switzerland), and processing there too, to eligible API customers and new ChatGPT Enterprise and Edu customers. Account data, billing, metadata and connected apps stay out.

Data leaving the EU needs a legal basis under Chapter V of the GDPR. For the US, the Commission’s decision of 10 July 2023 on the EU-US Data Privacy Framework covers only companies on the Data Privacy Framework List; otherwise the usual route is the Commission’s standard data protection clauses, one of the safeguards in Article 46.

Should the model run on your own server? Usually what you need is that the system around it does. Anthropic’s self-hosted Managed Agents (in private beta since 12 November 2026) run the agent, its tools and its session history in your own cloud or on your premises; model requests go where you choose.

No. 04

You stay responsible: controller, processor, contract

Under the GDPR your company is usually the controller: it decides why and how personal data is processed (Article 4). A provider that processes it on your behalf is a processor. Article 28 allows only processors with sufficient guarantees, under a contract that binds them to your documented instructions (transfers included), to your authorisation for sub-processors, and to delete or return the data at the end.

That contract is the data processing agreement, and the big three offer one for business use: Anthropic’s commercial terms include it and say it “may not train models on Customer Content”; OpenAI signs one for ChatGPT Business, Enterprise and the API; Google applies its own to the paid Gemini API. Check that the account your team actually uses is one of these.

No. 05

What to keep away, or limit

The GDPR asks for only the data that is necessary (Article 5), restricts health data and the other special categories (Article 9), and lets each country set conditions for national ID numbers (Article 87). In practice:

  • The agent reads only what its job needs. Answering about bookings does not need the ID documents folder.
  • ID documents, health data and bank details stay in your system, not in prompts or anyone’s personal drive.
  • Records you must keep unchanged for years stay in your system of record. The agent reads copies; it is never the archive.
No. 06

Staff, and owners, on personal accounts

The risk we hear about most is someone pasting a client file into a personal account to save ten minutes. A one-page rule fixes most of it: which tools, on which accounts, and what never goes in. The Spanish data protection authority (AEPD) said the same in its January 2026 recommendations: follow your organisation’s policies and keep out confidential data about the company, its staff and its clients.

The EU AI Act points the same way. Article 4, as amended by Regulation (EU) 2026/1744 of 8 July 2026, requires businesses using AI systems at work to take measures to support their staff’s AI literacy, without guaranteeing a specific level.

No. 07

When an agent acts for you

On 27 October 2026 the AEPD published a decalogue on delegating tasks to AI assistants and agents. It asks five questions first: what the agent can see, what it can remember, what it can do in your name, where it sends the data, and whether it affects anyone else. It recommends minimum permissions and confirmation before sending, sharing, paying or signing up, and warns that a web page, an email, a document or a calendar invitation can contain hidden instructions, so the best protection is to limit what the agent can do if it is fooled.

That is how we build: an agent that answers customers on WhatsApp reads only the records it needs, and anything outside its limits goes to a person first.

No. 08

Questions to ask any AI provider or studio

  1. Which plan or API, under which terms? Is our data used for training?
  2. How long are prompts and outputs kept? Is zero data retention available for our models?
  3. Where is the data stored, and where is the model run? What is outside that promise?
  4. Is there a data processing agreement, and who are the sub-processors?
  5. If data leaves the EU, on what basis: the Data Privacy Framework List or standard clauses?
  6. What can the agent read and do without a person, and how do we stop it?
  7. What is deleted or returned when we stop?

If you keep one thing: the plan decides what happens to your data, not the brand. Use business terms with a data processing agreement, give the agent only what it needs, and get the answers on location in writing.

Frequently asked questions

Questions about AI and your company’s data

Does ChatGPT or Claude train on what my team pastes in?
On personal accounts it can: OpenAI unless the user opts out, Anthropic if the user allows it. On their business plans and APIs, neither trains on your data by default.
Can Claude keep our data in the EU?
Yes, through regional endpoints on Amazon Bedrock or Google Cloud’s Vertex AI. On Anthropic’s own API, as of 19 November 2026, the model runs in the US or with global routing, and data at rest is stored in the US.
Should the AI model run on our own server?
Usually it is enough that the agent, its tools and its records run in your cloud, and that model requests go to an endpoint whose location and retention you agreed in writing.

Sources, checked on 19 November 2026: Anthropic, consumer training, consumer retention, commercial training, commercial retention, covered models, data usage, data residency, regional compliance, self-hosted Managed Agents (in private beta since 12 November 2026), commercial terms; OpenAI, how your data is used (13 March 2026), enterprise privacy, API data controls, data residency; Google, Gemini Apps privacy hub (17 November 2026), Workspace privacy hub, Gemini API terms (23 March 2026); EUR-Lex, GDPR Articles 4, 5, 9, 28, 44 to 46 and 87, Decision (EU) 2023/1795, Regulation (EU) 2026/1744; AEPD, decalogue on AI assistants and agents (27 October 2026) and recommendations on using AI (27 January 2026). This guide is not legal advice.

Keep reading
Your data · your rules

Not sure where your data would go?

Tell us what the agent should read and do. We will tell you plainly which plan, which region and which contract it needs, and what it should never see.